if you are building a classifier for tumor detection where the consequences of not correctly identifying a positive example can be high. what evaluation metric would you want to maximize?